Fig. 2

microCLIP in silico framework. a Dataset collection and methodology for positive and negative MRE identification. More than 6000 interactions were retrieved from direct techniques and miRNA-target chimeric fragments. Numerous high-throughput experimental data following specific miRNA perturbations enabled the identification of AGO bound or differentially transcribed/translated genes harboring functional binding sites. In order to resolve the exact miRNA-binding sites, positive and negative instances were coupled with signal from 24 AGO-PAR-CLIP libraries. The negative set was enhanced by incorporating background CLIP-Seq clusters. sRNA-Seq datasets were included to determine expressed miRNAs and accurately extract positive/negative MREs. This dataset collection was processed to form the training/test sets of microCLIP deployment (Supplementary Tables 2-3), while 18 miRNA perturbation experiments were segregated (Supplementary Table 6) and introduced in the analyses of Figs. 5–7. b Separate subsets of the positive/negative miRNA interactions were used to train the distinct levels of the algorithm’s modeling. 9 base classifiers in the first layer comprise characteristic feature subsets (Supplementary Data 1) that assemble into the GBM meta-learner of the second layer. A super learning scheme is utilized in 8 of the 9 base nodes, weighing outputs from seven individual models. “Region features” node corresponds to an RF classification scheme and consists of CLIP-sequencing-derived features. Five base models (2–6) were designed for MRE-specific features: “Binding Vectors” describe the (un)paired positions along the miRNA/MRE hybrid; “Matches per miRNA/MRE domain” contain attributes of miRNA-target structure and sub-domains; “Duplex Features” include free energy, secondary structure, and AU base pairing features for miRNA and/or target; “Base pairing” encompasses composition descriptors of (un)paired nucleotides; “MRE general” incorporates general MRE-related descriptors. Three supplementary classifiers (“Feature Combination Set 1–3”) comprise unique combinations of features found in base nodes 1–6